Deviations from pre-pandemic lung cancer mortality trends in the United States: a 25-year counterfactual analysis (1999–2023)

Abstract

Background

The COVID-19 pandemic coincided with substantial disruptions to cancer care services worldwide. However, comprehensive population-level assessments of deviations from expected lung cancer mortality trends during the pandemic period using rigorous counterfactual methods remain limited, particularly regarding heterogeneity across demographic and geographic subgroups.

Methods

We performed a counterfactual analysis examining lung cancer mortality trends among United States adults aged 45 years and older (1999–2023) using CDC WONDER mortality data. Joinpoint regression characterized pre-pandemic trends (1999–2019) to project expected mortality rates for 2020–2023. Mortality gaps were calculated as differences between observed and expected age-adjusted rates across 66 demographic and geographic strata. Interrupted time series analysis evaluated pandemic-associated trend modifications. Population-weighted linear regression assessed dose-response relationships between state-level COVID-19 burden and lung cancer excess mortality.

Results

National age-adjusted lung cancer mortality rates declined from 156.03 per 100,000 in 1999 to 94.90 in 2019 (annual percent change: -4.56, 95% CI: -4.92 to -4.20). During the pandemic period, cumulative excess lung cancer mortality totaled 11.26 per 100,000, with statistically significant positive gaps emerging in 2021 (+ 3.67), 2022 (+ 2.77), and 2023 (+ 4.99). Non-Hispanic White populations exhibited three-fold greater 2023 lung cancer excess (+ 6.21) compared to Hispanic populations (+ 1.81). Age-stratified analysis revealed sixteen-fold gradients in lung cancer mortality gaps, from + 1.53 per 100,000 in the 45–54 years cohort to + 24.75 in those aged 85 years and older. State-level cumulative lung cancer mortality gaps ranged from − 40.72 to + 63.16 per 100,000. A significant positive ecological association was observed between state-level cumulative COVID-19 mortality and cumulative lung cancer mortality gaps (β = 0.056, 95% CI: 0.010–0.103, P = 0.022).

Conclusions

The COVID-19 pandemic was associated with deviations from expected lung cancer mortality trends, exhibiting substantial demographic and geographic heterogeneity. These patterns may reflect differences in healthcare system capacity across populations. Population-specific surveillance systems and infrastructure investments warrant consideration to maintain cancer care continuity during future public health emergencies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-026-26429-x.

Introduction

The COVID-19 pandemic precipitated an unprecedented disruption to cancer care delivery systems worldwide, creating cascading collateral health consequences that extend beyond direct viral morbidity and mortality [1, 2]. Lung cancer, representing the leading cause of cancer-related death globally and disproportionately affecting older populations with substantial comorbidity burdens, provides a sentinel context for examining pandemic-period disruptions and constraints in cancer care delivery [3, 4]. The convergence of overlapping clinical presentations between COVID-19 and lung cancer symptomatology, coupled with systematic healthcare resource reallocation toward pandemic response, created conditions for widespread diagnostic delays and treatment modifications [5, 6]. These disruptions threatened to reverse decades of mortality improvements achieved through advances in screening, surgical techniques, and targeted therapies, while potentially amplifying pre-existing disparities in cancer outcomes across geographic and demographic populations [7]. The pandemic thus presented an unprecedented natural experiment for evaluating healthcare system resilience and quantifying the collateral impact of crisis-driven care disruptions on cancer mortality trends [8, 9].

Despite mounting evidence of pandemic-related disruptions to cancer care processes, comprehensive population-level assessments of mortality outcomes remain limited, with existing studies predominantly focusing on short-term institutional metrics rather than rigorous counterfactual analyses of long-term epidemiological trends [10, 11]. Current literature lacks systematic quantification of excess cancer mortality during the pandemic period using mortality gap methodologies that account for pre-pandemic secular trends and incorporate statistical uncertainty [12, 13]. Furthermore, the differential impact of COVID-19 across demographic and geographic populations has not been comprehensively characterized at the national scale, particularly regarding potential amplification of existing disparities among racial and ethnic minorities and across distinct geographic regions [14, 15]. The heterogeneous distribution of healthcare infrastructure, specialist availability, and socioeconomic determinants across states and census regions suggests that pandemic-related disruptions may have manifested with substantial geographic and demographic variation [16, 17]. This analytical gap represents a critical knowledge deficit in understanding the true public health burden of pandemic-related collateral damage and identifying populations most vulnerable to healthcare system resilience failures during crisis periods.

To address these knowledge gaps, we conducted a comprehensive counterfactual analysis of lung cancer mortality trends in the United States (U.S.) from 1999 to 2023, employing Joinpoint regression methodology to quantify pandemic-related deviations from expected mortality trajectories. Our study provides a national-level assessment of lung cancer mortality deviations during 2020–2023 using statistical frameworks that estimate mortality gaps and corresponding confidence intervals across demographic, geographic, and socioeconomic strata. We hypothesized that the pandemic would be associated with significant disruptions to the long-term declining trends in lung cancer mortality, with disproportionate impacts observed across census regions and among racial and ethnic minority populations, reflecting differential healthcare system resilience and amplification of pre-existing structural inequalities in cancer care access and delivery.

Methods

Study design and data sources

Mortality data were obtained from the Centers for Disease Control and Prevention’s Wide-ranging Online Data for Epidemiologic Research (CDC WONDER) database [18]. We extracted death certificate records for all U.S. residents aged 45 years and older who died from lung and bronchus cancer (International Classification of Diseases, 10th Revision (ICD-10) codes C34.0-C34.9) between 1999 and 2023. Age-adjusted mortality rates (AAMR) were calculated using direct standardization to the 2000 U.S. standard population [19, 20].

The study cohort was restricted to individuals aged 45 years and older, as epidemiological evidence demonstrates that lung cancer incidence and mortality in individuals younger than 45 years constitute an exceedingly small proportion of total cases, with the vast majority of diagnoses occurring in older populations [3, 4]. This age restriction optimized statistical power while maintaining clinical relevance. Temporal analysis was stratified into baseline (1999–2019) and pandemic (2020–2023) periods to enable robust counterfactual inference [21]. The study was exempt from institutional review board approval because CDC WONDER contains de-identified, publicly available data.

Population stratification and variable definition

Data extraction was performed across 66 distinct demographic and geographic strata to examine health disparities comprehensively. Geographic stratification included U.S. Census regions (Northeast, Midwest, South, West), individual states (n = 50), and the District of Columbia. Demographic variables encompassed sex (male, female), age groups (45–54, 55–64, 65–74, 75–84, 85 years and older), and race/ethnicity categories (Non-Hispanic White, Non-Hispanic Black, Hispanic, Non-Hispanic Other). For each subgroup-year combination, we extracted the number of deaths, population denominators, crude mortality rates, AAMR, and corresponding standard errors from CDC WONDER. Age adjustment employed the direct method using the 2000 U.S. standard population as the reference, with mortality rates expressed per 100,000 population [20].

Statistical analysis

Analytical framework overview

Our analytical strategy comprised four components with distinct objectives: (1) Primary analysis using Joinpoint regression to characterize pre-pandemic trends and generate counterfactual mortality projections; (2) Secondary validation using interrupted time series analysis to independently assess trend modification; (3) Exploratory dose-response analysis examining ecological associations between state-level COVID-19 burden and lung cancer mortality deviations; and (4) Sensitivity analyses using alternative baseline periods and ARIMA models to assess robustness of primary findings. The Joinpoint approach was selected as the primary method because it is the established standard for cancer mortality trend analysis endorsed by the National Cancer Institute and accounts for the segmented declining trends characteristic of lung cancer mortality data.

Primary analysis: joinpoint regression and counterfactual prediction

Pre-pandemic mortality trends were characterized using Joinpoint regression analysis through the National Cancer Institute’s Joinpoint Regression Program version 5.1.0 [22]. For each demographic and geographic subgroup, we fitted models with a maximum of three joinpoints to the baseline period (1999–2019), applying the permutation test method with Bonferroni correction for multiple comparisons [23]. Annual percentage changes (APCs) with 95% confidence intervals (CI) were calculated for the final temporal segment using Monte Carlo permutation testing. Expected mortality rates for 2020–2023 were calculated using the formula:

Because Joinpoint regression models rates on the logarithmic scale, projected values are constrained to remain positive regardless of projection horizon, avoiding the negative mortality rate predictions that could arise from linear extrapolation of absolute rates. All projected values for 2020–2023 remained positive across all subgroups. Confidence intervals for expected rates were derived from the variance of APC estimates [24, 25].

Mortality gaps were quantified as the difference between observed and expected AAMR, with statistical significance assessed using two-sample z-tests incorporating both observed and predicted variance components. Standard errors for mortality gaps were computed using error propagation [26]:

Secondary analysis: interrupted time series analysis

We implemented interrupted time series (ITS) analysis as an independent validation approach to test for pandemic-associated trend modification. The intervention was specified as the first pandemic year, 2020, to align with the annual resolution of the mortality data [27]. For each stratum, we fitted segmented linear regression models:

, where represents the age-adjusted mortality rate at year , is a continuous time index from 1 (1999) to 25 (2023), is a binary indicator (0 for 1999–2019, 1 for 2020–2023), denotes time since the intervention (0 for 1999–2019, and t-2019 for 2020–2023, taking values 1–4). In this parameterization,, captures baseline trend slope, quantifies immediate level change at intervention, and measures post-intervention slope change [28, 29]. To address potential heteroscedasticity and serial correlation, we employed Newey-West heteroscedasticity and autocorrelation consistent standard errors with optimal lag selection [30]. When significant autocorrelation was detected, models were refitted using methods that account for serial correlation in the error terms [31–33]. Moving block bootstrap with 1,000 iterations provided distribution-free uncertainty quantification, with block length optimized to preserve temporal dependence structure [34].

Exploratory analysis: dose-response analysis

To examine ecological associations at the state level, we conducted analysis examining the relationship between COVID-19 burden and lung cancer mortality gaps. The exposure variable was cumulative age-adjusted COVID-19 mortality rate (2020–2023) extracted from CDC WONDER using ICD-10 code U07.1. The outcome variable was cumulative lung cancer mortality gap across the pandemic period for each state. We fitted population-weighted linear regression models with heteroscedasticity-robust standard errors using the Davidson-MacKinnon HC3 estimator:

denotes the cumulative lung cancer mortality gap for state over 2020–2023, and ​ denotes the cumulative age-adjusted COVID-19 mortality rate for state over 2020–2023. Population weighting addressed heteroscedasticity arising from varying state population sizes, with weights proportional to the square root of state population [35, 36]. Bootstrap validation employed bias-corrected and accelerated (BCa) confidence intervals with 10,000 replications [37].

Sensitivity analysis

To assess robustness of primary counterfactual projections, we conducted comprehensive sensitivity analysis using an alternative baseline period (1999–2018), testing whether inclusion of 2019 data materially affects trend estimates. Comparative assessment utilized Bland-Altman analysis for systematic bias detection, Lin’s concordance correlation coefficient for method agreement, and parallel analysis using autoregressive integrated moving average (ARIMA) models with automatic order selection via the Akaike Information Criterion [38–41]. Supplementary changepoint detection using structural break tests validated our a priori intervention point selection [42–44].

Multiple testing correction

Given the large number of subgroups analyzed (n = 66), we implemented the Benjamini-Hochberg false discovery rate (FDR) procedure to control the expected proportion of false discoveries at α = 0.05 [45]. The correction was applied separately to level change and slope change parameters across all subgroups, with Bonferroni-corrected p-values computed for family-wise error rate control [46]. Post-hoc power analysis was conducted using Cohen’s conventions for effect size interpretation [47].

All analyses were conducted using R version 4.5.1 with parallel processing capabilities. Statistical significance was assessed at α = 0.05 for individual tests, with FDR correction applied for multiple comparisons using the Benjamini-Hochberg procedure.

Results

Baseline lung cancer mortality trends (1999–2019)

AAMR among adults aged 45 years and older declined from 156.03 per 100,000 population in 1999 to 94.90 per 100,000 in 2019, representing an absolute reduction of 61.13 per 100,000 with 39.2% relative decline (Table 1). Joinpoint regression analysis identified the final trend segment spanning 2014–2019 (6 years) with an annual percent change (APC) of -4.56 (95% CI: -4.92 to -4.20). This final-segment APC served as the baseline for counterfactual mortality projections during 2020–2023. Males demonstrated greater absolute mortality reductions of 103.25 per 100,000 with 47.6% relative decline compared with females at 32.68 per 100,000 with 29.0% relative decline. Final-segment APCs were − 5.08 (95% CI: -5.47 to -4.69) for males and − 4.12 (95% CI: -4.49 to -3.75) for females. Among racial and ethnic groups, non-Hispanic Black individuals exhibited 45.1% relative decline (APC: -5.07, 95% CI: -5.47 to -4.67), Hispanic populations demonstrated 40.6% decline (APC: -4.36, 95% CI: -5.16 to -3.54), non-Hispanic White individuals showed 35.6% decline (APC: -4.32, 95% CI: -4.57 to -4.06), and non-Hispanic Other populations exhibited 31.6% decline (APC: -3.79, 95% CI: -4.42 to -3.16) (Table S1).

Table 1.

Baseline lung cancer mortality trends by population subgroup, United States, 1999–2019

Population Subgroup	AAMR 1999 per 100,000	AAMR 2019 per 100,000	Absolute Decline per 100,000	Percent Decline	Final Segment APC (95% CI)
Total Population	156.03	94.90	61.13	39.2%	-4.56 (-4.92, -4.20)
Sex
Female	112.67	80.00	32.68	29.0%	-4.12 (-4.49, -3.75)
Male	217.15	113.89	103.25	47.6%	-5.08 (-5.47, -4.69)
Race/Ethnicity
Hispanic	70.73	42.00	28.73	40.6%	-4.36 (-5.16, -3.54)
Non-Hispanic Black	183.73	100.79	82.94	45.1%	-5.07 (-5.47, -4.67)
Non-Hispanic Other	83.55	57.19	26.36	31.6%	-3.79 (-4.42, -3.16)
Non-Hispanic White	161.02	103.70	57.32	35.6%	-4.32 (-4.57, -4.06)
Age Group
45–54 years	31.24	13.63	17.61	56.4%	-9.80 (-10.47, -9.12)
55–64 years	123.33	64.76	58.57	47.5%	-4.00 (-4.43, -3.57)
65–74 years	281.29	144.26	137.03	48.7%	-5.37 (-5.85, -4.88)
75–84 years	361.89	260.78	101.11	27.9%	-4.19 (-4.60, -3.77)
85 + years	296.85	281.11	15.74	5.3%	-2.79 (-3.58, -1.99)
Geographic Region
Midwest	158.97	107.72	51.25	32.2%	-4.26 (-4.63, -3.90)
Northeast	147.73	89.45	58.28	39.4%	-4.53 (-5.00, -4.07)
South	170.10	103.77	66.33	39.0%	-4.39 (-4.76, -4.02)
West	136.88	72.19	64.69	47.3%	-5.44 (-6.14, -4.73)

Age-specific rates represent crude mortality rates within each age stratum. Final Segment APC refers to the last temporal segment identified by Joinpoint regression, which for the total population spanned 2014–2019 (6 years). Complete segment information for all subgroups is provided in Supplementary Table S1

Abbreviations: AAMR Age-adjusted mortality rate per 100,000 population, directly standardized to the 2000 US standard population (ages 45 years and older, truncated standard); APC Annual percent change, CI Confidence interval

National mortality gaps during the pandemic period (2020–2023)

Temporal patterns of mortality gaps across the pandemic period demonstrated heterogeneous trajectories across population subgroups (Fig. 1). At the national level, observed AAMR in 2020 was 90.40 per 100,000 (95% CI: 89.92–90.89) compared with expected AAMR of 90.57 per 100,000 (95% CI: 90.23–90.92), yielding a mortality gap of -0.17 per 100,000 (95% CI: -0.76 to 0.43, non-significant) (Table 2). Statistically significant positive gaps emerged beginning in 2021 at + 3.67 per 100,000 (95% CI: 2.85–4.49), followed by + 2.77 per 100,000 in 2022 (95% CI: 1.72–3.81), and + 4.99 per 100,000 in 2023 (95% CI: 3.71–6.26). The 2023 gap magnitude exceeded both 2021 and 2022 levels. Cumulative excess mortality burden across the four-year pandemic period totaled + 11.26 per 100,000.

Fig. 1.

Heatmaps show annual mortality gaps—observed minus Joinpoint-projected age-adjusted mortality rates—per 100,000 population by sex, race/ethnicity, age group, and census region. Colors are centered at zero (blue = negative, red = positive), and cell labels display yearly values with a “+” prefix for positive gaps

Table 2.

Observed and expected age-adjusted lung cancer mortality rates and mortality gaps, United States, 2020–2023

Population Subgroup	Year	Observed AAMR (95% CI)	Expected AAMR (95% CI)	Mortality Gap (95% CI)
Total Population	2020	90.40 (89.92–90.89)	90.57 (90.23–90.92)	-0.17 (-0.76 to 0.43)
	2021	90.11 (89.62–90.60)	86.44 (85.79–87.10)	+ 3.67 (2.85–4.49)
	2022	85.27 (84.80-85.73)	82.50 (81.57–83.44)	+ 2.77 (1.72–3.81)
	2023	83.72 (83.26–84.18)	78.74 (77.56–79.94)	+ 4.99 (3.71–6.26)
Sex
Male	2020	108.15 (107.35-108.95)	108.11 (107.67-108.56)	+ 0.04 (-0.88 to 0.95)
	2021	106.32 (105.52-107.12)	102.62 (101.78-103.47)	+ 3.70 (2.53–4.86)
	2022	99.96 (99.20-100.72)	97.41 (96.21–98.62)	+ 2.55 (1.13–3.98)
	2023	96.76 (96.02–97.50)	92.46 (90.95-94.00)	+ 4.30 (2.60–5.99)
Female	2020	76.42 (75.81–77.02)	76.70 (76.41-77.00)	-0.29 (-0.96 to 0.39)
	2021	77.33 (76.72–77.95)	73.55 (72.98–74.11)	+ 3.79 (2.95–4.62)
	2022	73.97 (73.38–74.55)	70.52 (69.71–71.34)	+ 3.45 (2.44–4.45)
	2023	73.34 (72.76–73.92)	67.61 (66.58–68.66)	+ 5.73 (4.53–6.92)
Race/Ethnicity
Non-Hispanic White	2020	99.31 (98.72–99.91)	99.23 (98.97–99.49)	+ 0.08 (-0.56 to 0.73)
	2021	99.67 (99.07-100.27)	94.95 (94.45–95.44)	+ 4.72 (3.94–5.51)
	2022	94.23 (93.66–94.81)	90.85 (90.14–91.56)	+ 3.39 (2.47–4.30)
	2023	93.14 (92.57–93.71)	86.93 (86.02–87.84)	+ 6.21 (5.14–7.29)
Non-Hispanic Black	2020	94.35 (92.75–95.94)	95.68 (95.28–96.08)	-1.33 (-2.98 to 0.32)
	2021	94.19 (92.57–95.80)	90.83 (90.07–91.59)	+ 3.36 (1.57–5.15)
	2022	90.97 (89.41–92.53)	86.22 (85.14–87.32)	+ 4.75 (2.85–6.65)
	2023	87.33 (85.82–88.84)	81.85 (80.49–83.24)	+ 5.48 (3.44–7.52)
Hispanic	2020	39.76 (38.69–40.84)	40.17 (39.83–40.51)	-0.41 (-1.54 to 0.72)
	2021	40.11 (39.03–41.19)	38.42 (37.78–39.08)	+ 1.69 (0.42–2.95)
	2022	38.56 (37.53–39.59)	36.75 (35.83–37.69)	+ 1.81 (0.42–3.21)
	2023	36.96 (35.97–37.95)	35.15 (33.98–36.36)	+ 1.81 (0.26–3.36)
Non-Hispanic Other	2020	55.06 (53.46–56.66)	55.02 (54.66–55.38)	+ 0.04 (-1.60 to 1.68)
	2021	53.26 (51.77–54.75)	52.94 (52.25–53.64)	+ 0.32 (-1.32 to 1.96)
	2022	51.38 (49.97–52.78)	50.93 (49.94–51.94)	+ 0.45 (-1.28 to 2.17)
	2023	50.46 (49.10-51.83)	49.00 (47.73–50.30)	+ 1.46 (-0.41 to 3.34)
Age Group
45–54 years	2020	12.51 (12.17–12.86)	12.30 (12.21–12.39)	+ 0.21 (-0.14 to 0.57)
	2021	11.91 (11.58–12.25)	11.09 (10.93–11.26)	+ 0.82 (0.45–1.20)
	2022	11.08 (10.76–11.41)	10.01 (9.78–10.23)	+ 1.08 (0.68–1.47)
	2023	10.55 (10.24–10.87)	9.03 (8.76–9.30)	+ 1.53 (1.11–1.94)
55–64 years	2020	62.83 (62.08–63.59)	62.17 (61.90-62.45)	+ 0.66 (-0.14 to 1.46)
	2021	60.47 (59.74–61.21)	59.69 (59.16–60.23)	+ 0.78 (-0.13 to 1.69)
	2022	57.92 (57.19–58.65)	57.30 (56.54–58.08)	+ 0.61 (-0.45 to 1.67)
	2023	55.52 (54.81–56.24)	55.01 (54.04–56.01)	+ 0.51 (-0.71 to 1.73)
65–74 years	2020	139.94 (138.66-141.23)	136.52 (135.83-137.22)	+ 3.42 (1.96–4.88)
	2021	134.45 (133.21-135.69)	129.20 (127.89-130.52)	+ 5.25 (3.44–7.06)
	2022	131.86 (130.63-133.08)	122.27 (120.41-124.15)	+ 9.59 (7.35–11.83)
	2023	128.58 (127.38-129.77)	115.70 (113.37-118.09)	+ 12.87 (10.23–15.52)
75–84 years	2020	246.10 (243.71–248.50)	249.87 (248.78-250.96)	-3.76 (-6.40–1.13)
	2021	248.76 (246.33-251.19)	239.40 (237.32-241.51)	+ 9.35 (6.15–12.56)
	2022	232.48 (230.22-234.73)	229.38 (226.39-232.41)	+ 3.09 (-0.66 to 6.85)
	2023	225.96 (223.78-228.13)	219.78 (215.97-223.65)	+ 6.18 (1.77–10.59)
85 + years	2020	262.10 (258.22-265.99)	273.27 (271.05-275.51)	-11.16 (-15.65–6.68)
	2021	289.25 (284.94-293.56)	265.65 (261.35-270.02)	+ 23.61 (17.49–29.72)
	2022	259.81 (255.89-263.73)	258.24 (251.99-264.64)	+ 1.57 (-5.87 to 9.01)
	2023	275.79 (271.65-279.92)	251.04 (242.97-259.37)	+ 24.75 (15.57–33.93)
Geographic Region
Northeast	2020	83.81 (82.71–84.92)	85.40 (84.98–85.81)	-1.58 (-2.76–0.40)
	2021	81.49 (80.41–82.58)	81.53 (80.74–82.32)	-0.03 (-1.38 to 1.31)
	2022	78.29 (77.24–79.33)	77.83 (76.71–78.97)	+ 0.46 (-1.08 to 2.00)
	2023	76.12 (75.10-77.15)	74.30 (72.87–75.76)	+ 1.82 (0.06–3.59)
Midwest	2020	103.08 (101.95-104.21)	103.13 (102.73-103.52)	-0.05 (-1.25 to 1.15)
	2021	103.62 (102.48-104.76)	98.73 (97.98–99.49)	+ 4.89 (3.52–6.26)
	2022	97.92 (96.83–99.01)	94.52 (93.44–95.62)	+ 3.39 (1.85–4.93)
	2023	97.24 (96.16–98.32)	90.49 (89.12–91.89)	+ 6.74 (4.99–8.50)
South	2020	99.39 (98.56-100.21)	99.22 (98.83–99.60)	+ 0.17 (-0.74 to 1.08)
	2021	99.09 (98.26–99.92)	94.86 (94.13–95.60)	+ 4.23 (3.12–5.34)
	2022	93.17 (92.38–93.95)	90.70 (89.66–91.76)	+ 2.46 (1.15–3.78)
	2023	91.63 (90.85–92.40)	86.72 (85.39–88.07)	+ 4.91 (3.36–6.45)
West	2020	68.79 (67.89–69.68)	68.26 (67.76–68.77)	+ 0.52 (-0.51 to 1.56)
	2021	69.35 (68.43–70.26)	64.55 (63.59–65.52)	+ 4.80 (3.47–6.13)
	2022	65.70 (64.84–66.57)	61.04 (59.69–62.42)	+ 4.67 (3.05–6.28)
	2023	63.80 (62.95–64.65)	57.72 (56.02–59.47)	+ 6.08 (4.16-8.00)

Abbreviations: AAMR Age-adjusted mortality rate, CI Confidence interval

Sex-stratified mortality gaps

Sex-specific temporal trajectories revealed divergent patterns throughout the pandemic period (Fig. 1). In 2023, females exhibited a mortality gap of + 5.73 per 100,000 (95% CI: 4.53–6.92) compared with + 4.30 per 100,000 (95% CI: 2.60–5.99) for males (Table 2; Fig. 2). The 95% CI overlapped substantially, indicating no statistically significant difference between sexes at α = 0.05. Both sexes demonstrated statistically significant gaps across the latter three years of the pandemic period. Males showed gaps of + 3.70 per 100,000 in 2021, + 2.55 per 100,000 in 2022, and + 4.30 per 100,000 in 2023. Females demonstrated gaps of + 3.79 per 100,000 in 2021, + 3.45 per 100,000 in 2022, and + 5.73 per 100,000 in 2023.

Fig. 2.

(Top) Forest plot of excess mortality rates (per 100,000 population) with 95% confidence intervals for sex, race/ethnicity, age groups, and census regions. Diamond markers denote results that meet the dual criterion of FDR < 0.05 and 95% CI excluding 0. (Bottom) State-level mortality gaps (per 100,000 population) with 95% confidence intervals, ordered by magnitude; the vertical dashed line indicates zero

Race/ethnicity-stratified mortality gaps

Racial and ethnic disparities evolved throughout the pandemic period with persistent differences across groups (Fig. 1). By 2023, non-Hispanic White individuals exhibited a mortality gap of + 6.21 per 100,000 (95% CI: 5.14–7.29), with observed AAMR of 93.14 per 100,000 versus expected 86.93 per 100,000 (Table 2; Fig. 2). Non-Hispanic Black populations demonstrated a gap of + 5.48 per 100,000 (95% CI: 3.44–7.52), with observed 87.33 versus expected 81.85 per 100,000. Hispanic populations exhibited a 2023 gap of + 1.81 per 100,000 (95% CI: 0.26–3.36), with observed 36.96 versus expected 35.15 per 100,000. This gap was significantly lower than both non-Hispanic White and non-Hispanic Black populations, as demonstrated by non-overlapping confidence intervals. The ratio of non-Hispanic White to Hispanic gap magnitude was 3.4-fold. Non-Hispanic Other populations showed a 2023 gap of + 1.46 per 100,000 (95% CI: -0.41 to 3.34, non-significant), with observed 50.46 versus expected 49.00 per 100,000.

Age-stratified mortality gaps

Age-specific mortality patterns revealed substantial gradients throughout the pandemic period. In 2023, gaps ranged from + 1.53 per 100,000 (95% CI: 0.28–2.78) in individuals aged 45–54 years to + 24.75 per 100,000 (95% CI: 15.57–33.93) in individuals aged 85 years and older, representing a 16.2-fold gradient (Table 2; Fig. 2). Intermediate age groups demonstrated gaps of + 0.51 per 100,000 for 55–64 years (95% CI: -0.71 to 1.73, non-significant), + 12.87 per 100,000 for 65–74 years (95% CI: 10.23–15.52), and + 6.18 per 100,000 for 75–84 years (95% CI: 1.77–10.59). The 55–64 years age group constituted the sole stratum demonstrating a non-significant 2023 gap, despite significant gaps in both the adjacent younger 45–54 years group and older 65–74 years group.

Regional analysis

Regional mortality gap patterns evolved dynamically across the pandemic period (Fig. 1). In 2020, the Northeast region demonstrated a statistically significant negative gap of -1.58 per 100,000 (95% CI: -2.76 to -0.40), while the South, Midwest, and West regions showed non-significant gaps (Table 2). By 2023, all four regions exhibited positive mortality gaps. The Midwest demonstrated a gap of + 6.74 per 100,000 (95% CI: 4.99–8.50), with observed AAMR 97.24 per 100,000 versus expected 90.49 per 100,000. The West region showed a gap of + 6.08 per 100,000 (95% CI: 4.16-8.00), with observed 63.80 versus expected 57.72 per 100,000. The South exhibited a gap of + 4.91 per 100,000 (95% CI: 3.36–6.45), with observed 91.63 versus expected 86.72 per 100,000. The Northeast maintained the lowest 2023 gap at + 1.82 per 100,000 (95% CI: 0.06–3.59), with observed 76.12 versus expected 74.30 per 100,000.

State-level mortality gaps

State-level analysis revealed substantial geographic heterogeneity in pandemic-period lung cancer mortality gaps. Cumulative four-year mortality gaps spanning 2020–2023 across 51 jurisdictions ranged from + 63.16 per 100,000 at the maximum to -40.72 per 100,000 at the minimum, spanning 103.88 per 100,000 in total range (Fig. 3, Figure S4, Table S2). Ten states demonstrated cumulative gaps exceeding + 25.00 per 100,000, with the highest values observed in Alaska (+ 63.16), New Hampshire (+ 43.12), West Virginia (+ 39.66), Missouri (+ 36.29), and Arizona (+ 34.95). Ten jurisdictions exhibited negative cumulative gaps, with the most pronounced reductions observed in South Dakota (-40.72), North Dakota (-31.03), Hawaii (-26.08), Maine (-23.94), and Mississippi (-22.57). Geographic analysis demonstrated three spatial patterns (Fig. 3): northern tier and New England states showed predominantly high positive cumulative gaps; Appalachian and lower Midwest states exhibited moderate-to-high positive gaps; and Upper Plains and selected Mountain states demonstrated negative or minimal cumulative gaps.

Fig. 3.

(Top) Choropleth map of cumulative mortality gaps (per 100,000 population) across the 50 U.S. states and the District of Columbia. (Bottom) Bar charts list states with the highest and lowest cumulative mortality gaps, with values displayed for direct comparison

Interrupted time series analysis

Model fit statistics across 66 subgroups ranged from R² = 0.579 to R² = 0.996 (median R² = 0.958) (Table S3). Regional analysis identified statistically significant slope deceleration exclusively in the Western U.S. (β₃ = +1.460 per 100,000 population, 95% CI: 0.865 to 2.054, FDR q < 0.001). Racial and ethnic stratification revealed significant slope changes in Non-Hispanic Black populations (β₃ = +1.732, 95% CI: 0.601 to 2.863, FDR q = 0.008) and Hispanic populations (β₃ = +0.470, 95% CI: 0.083 to 0.858, FDR q = 0.042). Sex-stratified analyses yielded β₃ = +1.237 (95% CI: 0.260 to 2.214, FDR q = 0.035) for males and β₃ = +0.484 (95% CI: -0.553 to 1.520, FDR q = 0.492) for females. State-level analyses identified statistically significant slope modifications in 22 of 51 jurisdictions. Age stratification revealed statistically significant modifications exclusively in the 65–74 years cohort (β₃ = +3.569, 95% CI: 2.958 to 4.179, FDR q < 0.001). ITS-derived slope changes showed a moderate positive Spearman correlation with Joinpoint-based cumulative mortality gaps (ρ = 0.245, p = 0.047) (Fig. 4).

Fig. 4.

Scatter plot comparing the ITS-derived slope change (β₃, per year) with Joinpoint-based cumulative mortality gaps (per 100,000 population; 2020–2023) across subgroups. The fitted trend (95% confidence band), Spearman’s ρ, p-value, and sample size are reported

Sensitivity and validation analyses

Sensitivity analysis using a truncated baseline (1999–2018) demonstrated substantial concordance with the primary analysis across all 66 population subgroups. Mortality gap estimates exhibited strong correlation (Pearson r = 0.802; Lin’s concordance coefficient = 0.757), with a mean difference of 1.55 per 100,000 (95% CI: 0.31–2.79) and 95% limits of agreement spanning − 8.50 to 11.61 per 100,000 (Fig. 5). All subgroups maintained overlapping confidence intervals between methods, with no reversals in statistical significance observed. Autoregressive integrated moving average models generated mortality gap estimates across 176 subgroup-year strata (2020–2023), demonstrating moderate inter-method correlation with Joinpoint regression projections (r = 0.693; intraclass correlation coefficient = 0.672; Lin’s concordance correlation coefficient = 0.671). Bland-Altman analysis revealed mean inter-method bias of − 1.50 per 100,000 population with 95% limits of agreement spanning − 11.84 to + 8.85 per 100,000 (Fig. 6). Model diagnostic evaluation across 44 subgroups demonstrated adequate temporal autocorrelation specifications, with 100% of models satisfying white noise residual assumptions (Table S4).

Fig. 5.

(Left) Bland–Altman plot showing the difference (Sensitivity – Primary) versus the average mortality gap (per 100,000 population), with mean bias and 95% limits of agreement. (Right) Method-agreement summary reporting Lin’s concordance correlation coefficient (CCC) and Pearson correlation (r)

Fig. 6.

A Scatter plot of ARIMA- versus Joinpoint-estimated mortality gaps (per 100,000 population) with fitted regression and identity (y = x) lines; correlation and agreement metrics are reported. B Bland–Altman plot showing method differences versus means, with mean bias and 95% limits of agreement

Dose-response relationship analysis

Across all 51 U.S. jurisdictions during 2020–2023, cumulative age-adjusted COVID-19 mortality rates ranged from 422.9 to 995.2 per 100,000 population (median, 666.3), while cumulative lung cancer mortality gaps ranged from − 30.1 to 105.4 per 100,000 (median, 40.6). Population-weighted linear regression demonstrated a statistically significant positive association between these measures (β = 0.056 per 100,000 per 100-unit increase in COVID-19 mortality; 95% CI, 0.010–0.103; P = 0.022), with the model explaining 8.7% of variance in state-level mortality gaps (R² = 0.087) (Fig. 7, Table S5). Sensitivity analyses confirmed the robustness of this dose-response relationship across alternative regression specifications.

Fig. 7.

Dose–response between state COVID-19 mortality and excess lung cancer mortality, 2020–2023. Points are states. X: cumulative age-adjusted COVID-19 deaths per 100,000; Y: excess lung cancer deaths per 100,000. Solid line: population-weighted OLS with 95% CI

Discussion

This comprehensive counterfactual analysis documents deviations from long-term lung cancer mortality trends during the COVID-19 pandemic period, characterized by delayed emergence of observed excess, pronounced demographic heterogeneity, and marked geographic variation. National cumulative excess mortality accumulated across the pandemic period, with peak excess occurring several years after initial screening disruptions rather than during maximal service suspension periods. Demographic disparities manifested through substantial differences between racial and ethnic groups, alongside considerable age gradients. Geographic heterogeneity spanned a wide range across states, with dose-response evidence linking regional COVID-19 burden to lung cancer excess. These convergent findings challenge simplified disruption models and necessitate nuanced frameworks for interpretation that account for differential vulnerability across populations.

These findings contribute to the growing body of evidence examining pandemic-related cancer care disruptions. Previous investigations have primarily focused on short-term screening volume reductions and simulation-based mortality projections [2]. The present study advances this literature by providing an empirical national assessment of observed mortality deviations using counterfactual methods spanning the pandemic period through late 2023 [48]. The temporal pattern whereby excess peaked several years post-pandemic onset al.igns with biological latency predictions from modeling studies but contrasts with immediate effects observed in acute respiratory disease mortality. The demographic findings regarding attenuated excess in Hispanic populations diverge from conventional disparity predictions while paralleling patterns documented in COVID-19 direct mortality analyses. Geographic heterogeneity magnitudes exceed those reported in prior single-state analyses, suggesting national-scale assessment captures infrastructure variation masked in regional studies. The modest explanatory power of the dose-response relationship indicates multicausal complexity consistent with established competing risks frameworks.

The observed mortality patterns can be understood through the framework of competing risks, where populations with dual vulnerabilities to COVID-19 and cancer experience opposing mortality forces [48]. Acute viral deaths may selectively remove individuals at highest risk for cancer mortality, a phenomenon termed mortality displacement or harvesting effect. This mechanism would predict short-term cancer mortality deficits in regions experiencing severe viral burden. Conversely, healthcare disruptions generate delayed excess mortality through diagnostic postponements, treatment modifications, and surveillance gaps that manifest after biological latency periods [7]. The temporal dynamics observed in this study suggest delayed biological progression dominated acute competing mortality effects, though multiple alternative mechanisms warrant consideration. Potential pathways include diagnostic delays causing stage migration, treatment modifications independent of diagnosis timing, post-diagnostic surveillance interruptions, and direct COVID-19 comorbidity effects. Current aggregate mortality data cannot definitively distinguish these pathways; clinical registry linkage studies examining treatment completion rates, stage distributions, and cause-specific mortality attribution represent critical evidence gaps requiring systematic investigation [49].

The observed mortality patterns should be interpreted in the context of pandemic-related disruptions to lung cancer diagnosis and screening. Studies documented substantial declines in cancer screening during 2020, with screening volumes declining by up to 80–90% during peak pandemic periods before partially recovering [1, 10]. Diagnostic delays may initially reduce recorded lung cancer deaths if patients die from other causes before cancer diagnosis, but subsequently increase mortality as delayed diagnoses lead to more advanced stage presentations with poorer prognosis. The temporal pattern observed in this study, with minimal excess in 2020 followed by accumulating excess in subsequent years, is consistent with this delayed effect hypothesis, though we cannot directly test this mechanism with aggregate mortality data. Future research linking screening utilization data with stage-specific incidence and mortality would help quantify the contribution of diagnostic disruptions to observed excess mortality patterns.

Death certificate classification during the pandemic period warrants consideration. When multiple conditions are present, cause-of-death assignment may vary by clinical context and coding practices, and acute infectious conditions may be more likely to be selected as the underlying cause of death in some circumstances, potentially leading to under-ascertainment of lung cancer as the underlying cause during periods of high transmission. Two competing mechanisms may contribute to the temporal patterns we observed: mortality displacement, whereby COVID-19 caused premature death among lung cancer patients who might otherwise have died from cancer within months [48], and delayed care effects, whereby healthcare disruptions delayed diagnosis and treatment with mortality consequences emerging after a latency period. The modest 2020 mortality gap followed by accumulating excess in subsequent years is consistent with both mechanisms operating simultaneously, with early displacement effects potentially offsetting—at least temporarily—larger delayed-care effects over time, and the observed patterns potentially being influenced by cause-of-death coding during peak transmission periods.

The attenuated excess mortality observed in Hispanic populations compared to non-Hispanic White populations represents a descriptive finding that warrants cautious interpretation given multiple competing explanations. We observed that Hispanic populations exhibited smaller mortality gaps; however, several mechanisms may contribute to this pattern, and the ecological design cannot distinguish among them. First, lower baseline screening utilization may have limited absolute disruption magnitude through floor effects [50]. Second, younger age distributions may have reduced dual vulnerability to both COVID-19 and cancer mortality [51]. Third, stronger familial support networks may have facilitated care continuity. However, alternative explanations related to data limitations deserve equal consideration. Selective return migration, whereby seriously ill individuals return to countries of origin, may result in deaths occurring outside U.S. vital statistics capture [52]. Additionally, mortality underascertainment in immigrant populations due to incomplete death registration could contribute to apparently lower excess. Death certificate classification practices may also differ across populations. Importantly, interpreting lower observed excess as genuine healthcare system performance would be premature given these methodological uncertainties. Hispanic populations continue to face substantial baseline disparities in cancer care access, and these findings should not diminish attention to persistent structural barriers.

Geographic heterogeneity in observed mortality gaps requires acknowledgment of multiple competing explanations that cannot be distinguished with ecological data. We documented substantial variation across states; however, interpreting these patterns causally would exceed the evidence. Jurisdictions experiencing severe viral waves may have experienced selective depletion of populations with overlapping cancer vulnerability, potentially creating subsequent cancer mortality deficits [48]. Jurisdictions with lower baseline screening prevalence may have experienced proportionally smaller absolute declines when services were disrupted [53]. Coding practices may vary across jurisdictions, with dual-pathology deaths potentially assigned differentially to viral versus cancer causes depending on local conventions and healthcare system strain. Interstate migration patterns during the pandemic may have affected population denominators and mortality attribution. Completeness of death registration and cause-of-death ascertainment may also vary geographically. Current aggregate mortality data cannot distinguish among these mechanisms. Validation studies employing individual-level linkage between cancer registries and death certificates would help clarify which mechanisms predominate in different geographic contexts.

The heterogeneous patterns observed across populations suggest several considerations for public health practice, though specific policy recommendations require validation through additional research. First, the geographic variation in mortality gaps suggests that infrastructure and capacity differences across regions may warrant attention. States with historically lower screening utilization might benefit from targeted outreach, though the effectiveness of specific interventions such as mobile screening programs requires empirical evaluation [54, 55]. Second, the temporal pattern whereby excess mortality emerged after initial disruption periods suggests that surveillance systems capable of detecting delayed effects may be valuable. Enhanced mortality monitoring with capacity to compare observed and expected trends could facilitate earlier identification of emerging excess. Third, the substantial unexplained variance in state-level dose-response analyses indicates that factors beyond COVID-19 burden contribute to geographic heterogeneity. Future research should examine healthcare system characteristics, policy responses, and population factors that may explain differential patterns [56]. These observations should be interpreted as hypothesis-generating rather than definitive policy guidance, given the ecological design and multiple competing explanations for observed patterns.

Several research priorities emerge from these findings. First, cancer registry linkage studies examining individual-level screening completion, diagnostic stage distributions, and treatment timing would validate assumed mechanisms linking healthcare disruptions to observed mortality patterns. Second, natural experiment designs leveraging state-level policy heterogeneity could strengthen causal inference through quasi-experimental frameworks [57]. Third, multiple-cause mortality analyses with clinical validation are needed to quantify death certificate classification effects on observed patterns [58].

Several methodological limitations warrant consideration. Our counterfactual projections rely on the assumption that pre-pandemic trends would have continued through 2020–2023. The final Joinpoint segment spanned six years (2014–2019) with stable decline, supporting short-term extrapolation; however, if underlying trends would have naturally decelerated or accelerated, our estimates may over- or understate excess mortality. Sensitivity analyses using truncated baselines and ARIMA models showed substantial concordance, supporting robustness, but cannot fully resolve this inherent uncertainty. First, ecological design precludes individual-level causal inference, as unmeasured confounding including smoking trends, immunotherapy adoption, and treatment modification heterogeneity may partially explain observed associations. We did not directly analyze lung cancer incidence trends, though diagnostic disruptions represent an important hypothesized mechanism warranting dedicated investigation using cancer registry data. Second, death certificate classification introduces potential bias in both directions, whereby viral overcoding may mask cancer contributions while undercoding may inflate observed cancer excess. Third, stage distribution data validating assumed diagnostic delay mechanisms were unavailable in this analysis. Fourth, the modest variance explanation in dose-response analyses suggests multicausal complexity or potential measurement error. Fifth, methodological artifacts in Hispanic populations including selective return migration and mortality underascertainment may influence observed patterns. These limitations underscore the importance of cautious interpretation and highlight priorities for validation studies linking cancer registries with medical records to examine individual-level care trajectories. Additionally, our annual time resolution may obscure within-year dynamics, though monthly data would introduce greater stochastic variation and lung cancer mortality does not respond immediately to healthcare disruptions given biological latency periods.

Conclusion

This counterfactual analysis documents deviations from long-term lung cancer mortality trends during the COVID-19 pandemic period, with delayed emergence of observed excess and substantial demographic and geographic heterogeneity. These patterns may reflect differences in healthcare system capacity across populations, though the ecological design precludes causal inference regarding specific mechanisms. The findings underscore the potential value of population-specific surveillance systems and infrastructure investments to maintain cancer care continuity during future public health emergencies.

Author’s contributions

SC designed the research, collected, analyzed the data, drafted the manuscript, and revised the manuscript.

Funding

This work did not receive a specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

All data is publicly available at https://wonder.cdc.gov/.

Declarations

Ethics approval and consent to participate

Approval from an institutional review board was not required as the data analyzed was deidentified and in a publicly available database.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.